Deep Reinforcement Learning Approach for Joint Pricing and Charging in AEV Ride-Hailing Systems

Abstract

Ride-hailing services using autonomous electric vehicles (AEVs) at transportation hubs face strong demand fluctuations and limited charging resources. During peak periods, pricing decisions directly affect passenger demand, while charging decisions determine vehicle availability. Treating these two decisions independently often results in inefficient operations under highly dynamic conditions. Therefore, a coordinated decision-making approach is required. This study proposes a joint pricing and charging decision framework for AEV ride-hailing systems at transportation hubs based on deep reinforcement learning (DRL). A simulation environment is developed to represent passenger arrivals, AEV state-of-charge dynamics, and charging station constraints. A proximal policy optimization (PPO)–based model, named JPCPPO, is designed to jointly determine pricing levels and charging thresholds at each decision stage using aggregated system state information. Through continuous interaction with the environment, the model learns adaptive strategies for demand regulation and fleet energy management under time-varying conditions. Simulation results under different environment scales show that the proposed JPCPPO model consistently outperforms fixed pricing strategies, heuristic methods, and conventional actor-critic baselines in terms of overall reward and service performance. The additional experiments indicate that the computational cost of model training remains stable as the environment scale increases, demonstrating the applicability of the proposed approach to large-scale transportation hub operations.